In a recent post, we used demographic data and prior election results from Virginia and South Carolina to predict North Carolina’s primary election. Like most predictions, we got some right, and some very wrong. In this post, we review why we were right on some things and wrong on others. In doing so, we hope to show how election trends—including Trump’s and Clinton’s consistent areas of strength, and how establishment Republican support for Cruz—played out in North Carolina.
What we got (mostly) right
North Carolina held its Democratic and Republican primaries on March 15th. Donald Trump won the Republican race with 40% of the vote (see Map #1), while on the Democratic side Hillary Clinton won with 55% (see Map #2)
Despite our model’s limitations, we correctly predicted Trump’s victory, and our prediction (38%) was close to his actual vote total (40%). We also correctly predicted that Trump would win a majority of the state’s counties, and that his support would be concentrated away from the major metropolitan areas of the Triangle and Triad.
We also correctly predicted that Hillary Clinton would win the Democratic vote in North Carolina, and that she would carry a majority of the state’s counties. However, Bernie Sanders exceeded our state-wide predictions by five percentage points, and he polled especially well in the state’s far western counties, in addition to heavily-African American areas in northeastern North Carolina (see Map #3). Despite exceeding projections in the northeast part of the state, however, Sanders still lost most of those counties handily.
What we got (very) wrong
In a word: Cruz.
While we predicted that Cruz would receive 20% of the statewide vote and place third, he actually polled at 36% and came in second. Cruz’s support was the toughest to predict in the model,* so we didn’t feel especially confident in our projections compared to, say, Trump. We also knew that Rubio was not going to perform as well as he did in South Carolina or Virginia. What we didn’t know, however, was who voters the model projected to vote for Rubio would support.
As discussions in the news media at the time centered on the zero-sum, prisoner’s dilemma facing Rubio and Kasich (whose supporters mirrored each other nearly perfectly), we presented two hypotheticals: if one left the race (Rubio, for example) and all his supporters voted for the other (Kasich in this example), or if the two split their voters evenly.
While we purposefully skewed those scenarios to illustrate the prisoner’s dilemma, what happened instead was that a significant portion of Rubio-Kasich supporters voted for Cruz, especially in the Triangle (see Map #4). Rubio-Kasich underperformed their combined projections by over 25% in each of Wake, Durham, Orange, and Granville Counties (see Map #5). Conversely, Cruz outperformed projections in those counties by 15-30%.
The consolidation of non-Trump support behind Cruz also affected which candidate won particular counties. Cruz won 22 counties, including some fairly-populous ones in the Triangle and Triad (Wake, Durham, Guilford, and Forsyth) and more rural counties in Eastern and Western North Carolina.
Looking back, moving forward
Comparing our projections for the North Carolina Republican primary to the actual results, we see two important takeaways. First, Trump’s support is both stable and relatively predictable. He does well in counties with lower rates of college attendance and high school degrees, and with higher rates of those receiving Social Security or public health insurance (Medicare, Medicaid), who are disabled, and who were born in the U.S.
Second, North Carolina may have proven a bellwether of establishment Republican support coalescing around Ted Cruz. The “establishment” candidates—Rubio and Kasich—significantly underperformed projections, while Cruz exceeded them—especially in the more urban and suburban counties of the Triangle.
On the Democratic side, Sanders showed stronger support among African-American voters, but he still trailed Clinton in heavily-African American counties in northeastern North Carolina. Sanders also performed better in Appalachian North Carolina than he did in our projections, though comparatively few people live there in comparison to the state’s large urban areas. Despite exceeding projections, though, Sanders failed to gain the decisive victory that would allow him to catch Clinton in the delegate count.
*Statistically-speaking, the r-squared of Cruz’s model was substantially lower than that of any other candidate.